In linguistics, the minimalist program is a major line of inquiry that has been developing inside generative grammar since the early 1990s, starting with a 1993 paper by Noam Chomsky.[1]
Following Imre Lakatos's distinction, Chomsky presents minimalism as a program, understood as a mode of inquiry that provides a conceptual framework which guides the development of linguistic theory. As such, it is characterized by a broad and diverse range of research directions. For Chomsky, there are two basic minimalist questions—What is language? and Why does it have the properties it has?—but the answers to these two questions can be framed in any theory.[2]
See also: Biolinguistics.
Minimalism is an approach developed with the goal of understanding the nature of language. It models a speaker's knowledge of language as a computational system with one basic operation, namely Merge. Merge combines expressions taken from the lexicon in a successive fashion to generate representations that characterize I-Language, understood to be the internalized intensional knowledge state as represented in individual speakers. By hypothesis, I-language—also called universal grammar—corresponds to the initial state of the human language faculty in individual human development.
Minimalism is reductive in that it aims to identify which aspects of human language—as well the computational system that underlies it—are conceptually necessary. This is sometimes framed as questions relating to perfect design (Is the design of human language perfect?) and optimal computation (Is the computational system for human language optimal?) According to Chomsky, a human natural language is not optimal when judged based on how it functions, since it often contains ambiguities, garden paths, etc. However, it may be optimal for interaction with the systems that are internal to the mind.[3]
Such questions are informed by a set of background assumptions, some of which date back to the earliest stages of generative grammar:[4]
Minimalism develops the idea that human language ability is optimal in its design and exquisite in its organization, and that its inner workings conform to a very simple computation. On this view, universal grammar instantiates a perfect design in the sense that it contains only what is necessary. Minimalism further develops the notion of economy, which came to the fore in the early 1990s, though still peripheral to transformational grammar. Economy of derivation requires that movements (i.e., transformations) occur only if necessary, and specifically to satisfy to feature-checking, whereby an interpretable feature is matched with a corresponding uninterpretable feature. (See discussion of feature-checking below.) Economy of representation requires that grammatical structures exist for a purpose. The structure of a sentence should be no larger or more complex than required to satisfy constraints on grammaticality.
Within minimalism, economy—recast in terms of the strong minimalist thesis (SMT)—has acquired increased importance.[6] The 2016 book entitled Why Only Us—co-authored by Noam Chomsky and Robert Berwick—defines the strong minimalist thesis as follows:
Under the strong minimalist thesis, language is a product of inherited traits as developmentally enhanced through intersubjective communication and social exposure to individual languages (amongst other things). This reduces to a minimum the "innate" component (the genetically inherited component) of the language faculty, which has been criticized over many decades and is separate from the developmental psychology component.
Intrinsic to the syntactic model (e.g. the Y/T-model) is the fact that social and other factors play no role in the computation that takes place in narrow syntax; what Chomsky, Hauser and Fitch refer to as faculty of language in the narrow sense (FLN), as distinct from faculty of language in the broad sense (FLB). Thus, narrow syntax only concerns itself with interface requirements, also called legibility conditions. SMT can be restated as follows: syntax, narrowly defined, is a product of the requirements of the interfaces and nothing else. This is what is meant by "Language is an optimal solution to legibility conditions" (Chomsky 2001:96).
Interface requirements force deletion of features that are uninterpretable at a particular interface, a necessary consequence of Full Interpretation. A PF object must only consist of features that are interpretable at the articulatory-perceptual (A-P) interface; likewise a LF object must consist of features that are interpretable at the conceptual-intentional (C-I) interface. The presence of an uninterpretable feature at either interface will cause the derivation to crash.
Narrow syntax proceeds as a set of operations—Merge, Move and Agree—carried out upon a numeration (a selection of features, words etc., from the lexicon) with the sole aim of removing all uninterpretable features before being sent via Spell-Out to the A-P and C-I interfaces. The result of these operations is a hierarchical syntactic structure that captures the relationships between the component features.
The exploration of minimalist questions has led to several radical changes in the technical apparatus of transformational generative grammatical theory. Some of the most important are:[7]
Early versions of minimalism posits two basic operations: Merge and Move. Earlier theories of grammar—as well as early minimalist analyses—treat phrasal and movement dependencies differently than current minimalist analyses. In the latter, Merge and Move are different outputs of a single operation. Merge of two syntactic objects (SOs) is called "external Merge". As for Move, it is defined as an instance of "internal Merge", and involves the re-merge of an already merged SO with another SO.[8] In regards to how Move should be formulated, there continues to be active debate about this, but the differences between current proposals are relatively minute.
More recent versions of minimalism recognize three operations: Merge (i.e. external Merge), Move (i.e. internal Merge), and Agree. The emergence of Agree as a basic operation is related to the mechanism which forces movement, which is mediated by feature-checking.
See also: Merge (linguistics). In its original formulation, Merge is a function that takes two objects (α and β) and merges them into an unordered set with a label, either α or β. In more recent treatments, the possibility of the derived syntactic object being un-labelled is also considered; this is called "simple Merge" (see Label section).
In the version of Merge which generates a label, the label identifies the properties of the phrase. Merge will always occur between two syntactic objects: a head and a non-head.[9] For example, Merge can combine the two lexical items drink and water to generate drink water. In the Minimalist Program, the phrase is identified with a label. In the case of drink water, the label is drink since the phrase acts as a verb. This can be represented in a typical syntax tree as follows, with the name of the derived syntactic object (SO) determined either by the lexical item (LI) itself, or by the category label of the LI:
- | Merge (drink, water) → | Merge (drinkV, waterN) → |
Chomsky's earlier work defines each lexical item as a syntactic object that is associated with both categorical features and selectional features.[10] Features—more precisely formal features—participate in feature-checking, which takes as input two expressions that share the same feature, and checks them off against each other in a certain domain.[11] In some but not all versions of minimalism, projection of selectional features proceeds via feature-checking, as required by locality of selection:[12] [13] [14]
Selection as projection: As illustrated in the bare phrase structure tree for the sentence The girl ate the food; a notable feature is the absence of distinct labels (see Labels below). Relative to Merge, the selectional features of a lexical item determine how it participates in Merge:
Feature-checking: When a feature is "checked", it is removed.
Locality of selection (LOS) is a principle that forces selectional features to participate in feature checking. LOS states that a selected element must combine with the head that selects it either as complement or specifier. Selection is local in the sense that there is a maximum distance that can occur between a head and what it selects: selection must be satisfied with the projection of the head.
Move arises via "internal Merge".
Movement as feature-checking: The original formulation of the extended projection principle states that clauses must contain a subject in the specifier position of spec TP/IP.[15] In the tree above, there is an EPP feature. This is a strong feature which forces re-Merge—which is also called internal merge—of the DP the girl. The EPP feature in the tree above is a subscript to the T head, which indicates that T needs a subject in its specifier position. This causes the movement of
A substantial body of literature in the minimalist tradition focuses on how a phrase receives a proper label.[16] The debate about labeling reflects the deeper aspirations of the minimalist program, which is to remove all redundant elements in favour of the simplest analysis possible.[17] While earlier proposals focus on how to distinguish adjunction from substitution via labeling, more recent proposals attempt to eliminate labeling altogether, but they have not been universally accepted.
Adjunction and substitution: Chomsky's 1995 monograph entitled The Minimalist Program outlines two methods of forming structure: adjunction and substitution. The standard properties of segments, categories, adjuncts, and specifiers are easily constructed. In the general form of a structured tree for adjunction and substitution, α is an adjunct to X, and α is substituted into SPEC, X position. α can raise to aim for the Xmax position, and it builds a new position that can either be adjoined to [Y-X] or is SPEC, X, in which it is termed the 'target'. At the bottom of the tree, the minimal domain includes SPEC Y and Z along with a new position formed by the raising of α which is either contained within Z, or is Z.[18] Adjunction: Before the introduction of bare phrase structure, adjuncts did not alter information about bar-level, category information, or the target's (located in the adjoined structure) head.[19] An example of adjunction using the X-bar theory notation is given below for the sentence Luna bought the purse yesterday. Observe that the adverbial modifier yesterday is sister to VP and dominated by VP. Thus, the addition of the modifier does not change information about the bar-level: in this case the maximal projection VP. In the minimalist program, adjuncts are argued to exhibit a different, perhaps more simplified, structure. Chomsky (1995) proposes that adjunction forms a two-segment object/category consisting of: (i) the head of a label; (ii) a different label from the head of the label. The label L is not considered a term in the structure that is formed because it is not identical to the head S, but it is derived from it in an irrelevant way. If α adjoins to S, and S projects, then the structure that results is L =, where the entire structure is replaced with the head S, as well as what the structure contains. The head is what projects, so it can itself be the label or can determine the label irrelevantly. In the new account developed in bare phrase structure, the properties of the head are no longer preserved in adjunction structures, as the attachment of an adjunct to a particular XP following adjunction is non-maximal, as shown in the figure below that illustrates adjunction in BPS. Such an account is applicable to XPs that are related to multiple adjunction.[20]
AdjunctionDEFINITION: <H(S), H (S)>, where Label = (S = head)
- | Adjunction in X-bar theory | Adjunction in bare phrase structure |
SubstitutionDEFINITION: Label =